Quiz: AI Data Center Operations & Lifecycle (Part 14)

The top-line metric a mature AI fleet is managed against is…

facility availability (the nines)goodput — the fraction of bought GPU-hours converted to useful workPUE aloneticket closure rate

The DCIM/telemetry decision that makes failures diagnosable is…

buying more dashboardslanding facility (BMS/SCADA) and IT (DCGM/Redfish) data in one correlated time-series planelonger retentionalert de-duplication only

The checkpoint tier mix exists to buy down…

storage costmean time to recovery — RAM/peer tiers restore in seconds-to-minutes vs the persistent tier's longer pathmodel qualitynetwork bandwidth

Spares depth per failure class is set from…

a fixed 5% rulethe component's annualized failure rate × the replenishment lead timewhatever fits the storeroomvendor recommendations only

How much power-oversubscription headroom does synchronous training leave?

~50% — plenty~21% like inference~3% — training's synchronized peaks leave almost nothingunlimited with software caps

Firmware across the fleet (BMC, GPU, NIC, NVSwitch, PSU) should be treated as…

set-and-forgetpart of the attested golden stack, upgraded as a coordinated fleet event with canarieseach vendor's auto-updater's jobirrelevant to reliability

What triggers a refresh for a fleet slice, in the guide's framing?

the calendar alonecalendar, efficiency (perf-per-watt vs the power budget), or revenue signals — an explicit trigger you pick per slicewhen the vendor launches marketingfailure of the whole fleet

The facility's end state should be decided as…

demolition, alwaysrepower vs demolish vs remediate — with repower keeping the shell and the hard-won megawatts for the next buildabandonment in placeconversion to offices

The 'autonomy ladder' for facility control loops runs…

manual → automaticobserve → recommend → act-within-envelope → fully autonomous, chosen per control loopoff → onlocal → cloud

Disciplined operation on a live plant means…

freezing all changes foreverevery live-plant action gated by an approved, peer-reviewed procedure — human-error control as a designed systemtrusting experienced techs' judgmentoutsourcing operations

Check your understanding: Day-2 Operations, Upgrades & Lifecycle